Export For Second Brain Prompt

# Export-for-Second-Brain Prompt

> **For the operator.** `/crystallize` captures reasoning that happened **inside a Claude Code session**. This file is the other half: a prompt you paste into **Claude.ai (web)**, **ChatGPT**, **Gemini**, or **NotebookLM** to extract a conversation or notebook into a single markdown file, shaped so `/ingest` can file it cleanly.

## When to use this

- You had a long Claude.ai / ChatGPT / Gemini conversation that produced durable insight and you want it in the vault.
- A NotebookLM notebook has synthesised something you want to preserve as a source.
- You're migrating historical chat threads from one platform into the vault.

**When not to use.** Trivial back-and-forth, one-shot Q&A lookups, anything secret (health, credentials, financial). Those don't belong in the vault regardless of platform.

## How to use this

1. Open the third-party platform where the conversation lives.
2. Make sure the full conversation is loaded in the UI (scroll to the top if the platform lazy-loads).
3. Copy the block between `===== BEGIN PROMPT =====` and `===== END PROMPT =====` below and paste it as your next message into that platform.
4. The platform will reply with a single markdown document.
5. **Save the reply.** Copy the entire markdown output into a new file at `~/second-brain/inbox/<platform>-<short-slug>-YYYY-MM-DD.md`. Example: `inbox/chatgpt-pricing-model-debate-2026-04-23.md`.
6. In Claude Code (from the vault root): `/ingest inbox/`. The ingest skill will process the file like any other source — finding the wiki pages it touches, extracting claims, filing the raw under `raw/conversations/YYYY-MM/`, committing.

**NotebookLM note.** NotebookLM isn't a chat — it's a notebook with sources and synthesised answers. Paste the prompt into the notebook's chat pane; it will export the notebook's synthesised knowledge plus the source citations. The resulting markdown lists the underlying sources so `/ingest` can flag which raw documents you may want to pull in separately.

---

===== BEGIN PROMPT =====

I need you to export the full substantive content of this conversation (or, if this is a NotebookLM notebook, the notebook's synthesised knowledge and source inventory) into a single markdown document that I will file in a long-lived personal knowledge vault. Another system will ingest your output — so the format below is a **contract**, not a suggestion. Follow it exactly.

## Invariants (do not violate)

1. **Preserve substantive detail.** This is a high-fidelity export, not an executive summary. A 90-minute conversation should compile to 120–250 lines of markdown. If you strip out a decision, a framework, a named entity, a number, a rationale, a verbatim turn-of-phrase that carries reasoning — you've failed. The downstream vault explicitly rejects shallow captures.
2. **Do not invent content.** If a section below has no material in the conversation, write `_(none in this conversation)_` and move on. Fabrication, paraphrase-drift, or adding plausible-sounding content I didn't say is the single worst failure mode. If you're unsure whether something was actually said vs implied, prefer `_(implied but not stated: …)_` over silently asserting it.
3. **Attribute.** When a specific claim came from me (the user), tag it `[user]`. When it came from you (the assistant), tag it `[assistant]`. When it came from a cited source inside the conversation (a paper, article, link I pasted), tag it `[source: <title or URL>]`. Unattributed claims are worse than missing ones.
4. **Preserve numbers and names verbatim.** Entities, people, org names, product names, metrics, dates, €/$ figures, percentages, code identifiers — copy as they appeared. Do not round, translate, or normalise.
5. **Output a single markdown document.** No preamble, no "Here's your export:", no trailing commentary. The document starts with `---` (YAML frontmatter) and ends with the last content line. I will copy your entire reply verbatim into a file.

## Output format — follow this exactly